Walking through RMarkdown

We will examine the file data/college_data.csv. The file contains more than 160 different variables for the 395 doctoral universities in the US. The data was obtained from Integrated Postsecondary Education Data System (IPEDS). The variables correspond the most recent available academic year. In most of the cases, this mean 2018-19. However, some variables are from the 2017-18 year.

Setup

• We will include a large number of pictures.
• We don’t want to specify for every single code chunk to center and stretch-to-fit every plot.
• Thus we set those global options from beforehand.

knitr::opts_chunk$set(fig.align='center')
knitr::opts_chunk$set(out.width='100%')

• Next we load the data and our plotting functions from utils.R.

source('utils/utils.R')

## 
## Attaching package: 'plotly'

## The following object is masked from 'package:ggplot2':
## 
##     last_plot

## The following object is masked from 'package:stats':
## 
##     filter

## The following object is masked from 'package:graphics':
## 
##     layout

data <- read.csv('data/college_data.csv')
ID <- "Institution.Name" 
SZ <- "Total..enrollment..DRVEF2018." 
carnegie <- "Carnegie.Classification..HD2018."
landgrant <- "Land.Grant.Institution..HDNo0Yes8."
sector <- "Sector.of.institution..HDPrivate0Public8."

• We will later be able to identify every data point with its institutional name and its total enrollment.
• The plots can distinguish using two out of three criteria:
  • carnegie: Carnegie Classification as of 2018:
    • Very High Research
    • High Research
    • Research
  • landgrant: If it is a landgrant institution or not.
  • sector: Whether the institution is public or non-for-profit private.

Snapshot: Staff vs Faculty

Y <- "Total.FTE.staff..DRVHR2018."
X <- "Instructional..research.and.public.service.FTE..DRVHR2018."  

regression <- linregression_traditional(data,X,Y,carnegie)

summary_reg <- summary(regression)
rsquared <- summary_reg$r.squared
coeffs <- summary_reg$coefficients

• We observe an \(R^2\) value of 0.9072477, which is quite high
• For every research/instructional employee, we have 3 staff employees!

Use ggplot2 instead: nicer plots

foo <- ggplotRegression(data,Y,X,carnegie,sector,ID,SZ,'US Doctoral Universities')
foo[[1]]

summary_reg <- summary(foo[[2]])
rsquared <- summary_reg$r.squared
coeffs <- summary_reg$coefficients

• We observe an \(R^2\) value of 0.9072477, which is quite high
• For every research/instructional employee, we have 3 staff employees!

Use plotly to get an interactive plot!

• RMarkdown supports a number of different HTML widgets that breathe life into your data!
• HTML widgets are based on Javascript
  • Do not worry about it: R got your back.
  • plotly works well with ggplot2.

ggplotly(foo[[1]], tooltip = c('id', 'enroll', 'x', 'y'))

• Find more information on how to make interactive plots here.
• Might have issues when converting HTML widgets to PDF.
• Snapshot of widget based on phantomJS.

library(webshot)
webshot::install_phantomjs()

Look for other possible correlations

X <- "Undergraduate.enrollment..DRVEF2018."  
Y <- "Graduate.enrollment..DRVEF2018."  
foo <- ggplotRegression(data,Y,X,carnegie,sector,ID,SZ,'US Doctoral Universities')
foo[[1]]

summary_reg <- summary(foo[[2]])
rsquared <- summary_reg$r.squared
coeffs <- summary_reg$coefficients

• We observe an \(R^2\) value of 0.3357747, which is low.
• For every enrolled undergrad, we have 0 enrolled grad student.

Instead, let’s look at full time

X <- "Full.time.undergraduate.enrollment..DRVEF2018."  
Y <-  "Full.time.graduate.enrollment..DRVEF2018."  
foo <- ggplotRegression(data,Y,X,carnegie,sector,ID,SZ,'US Doctoral Universities')
foo[[1]]

summary_reg <- summary(foo[[2]])
rsquared <- summary_reg$r.squared
coeffs <- summary_reg$coefficients

• Considering full time enrollment doesn’t make the trend clearer.
• We observe an \(R^2\) value of 0.3332552, which remains low.
• For every enrolled undergrad, we have 0 enrolled grad student.

What about the number of degrees granted?

• Are Very High Research Activity institutions producing considerably more academic degrees than the rest of institutions?

p <- ggplotJitter(data, PhD, carnegie, sector, landgrant, ID, SZ)
ggplotly(p, tooltip = c('id', 'enroll', 'y'))

• The VHR institutions have a clear higher output
• But they also have a larger student body
• What if we normalize with respect to the number of total enrolled graduate students (we lack the specific number of enrolled PhD students)?
• Consider then the percentage \[ \frac{\#(\text{PhD degrees granted})}{\#(\text{Enrolled graduate students})}\times100 \]

df <- data.frame(
  data[c(carnegie,PhD,ID,SZ,sector,landgrant)],
  Students.receiving.a.PhD.normalized..DRVC2018. = as.vector(unlist(data[PhD]/data[GradT]))*100
)
Normalized <- "Students.receiving.a.PhD.normalized..DRVC2018."

p <- ggplotJitter(df, Normalized, carnegie, sector, landgrant, ID, SZ)
ggplotly(p, tooltip = c('id', 'enroll', 'y'))

Your Turn

• Plot other possible linear relationships.
  • Average price vs total enrollment.
  • Instructional vs Research FTEs.
• Examine bachelor and master output, both raw and normalized.